search

NVIDIA-AI-IOT / cuDLA-samples

YOLOv5 on Orin DLA
Other
185 stars 18 forks source link

Standalone mode and performance issue #7

Closed tanfengji-monarchtractor closed 9 months ago

tanfengji-monarchtractor commented 1 year ago

Hi, I'm using jetpack 5.1.2. The hybrid mode works fine, but the standalone mode returns the error below:

standalone_error

How may I fix this?

Besides, the performance is for hybrid mode is below:

yolov5s_hybrid_23ms

The MAP is 0.354 instead of 0.37+ mentioned in the README. Any possible reasons? Many thanks

zerollzeng commented 1 year ago

Sorry for the late response, please pull the latest code and rebuild with make validate_cudla_int8 USE_DLA_STANDALONE_MODE=1 USE_DETERMINISTIC_SEMAPHORE=1

Root cause in https://github.com/NVIDIA-AI-IOT/cuDLA-samples/issues/5

ou525 commented 10 months ago

image The latest code pulled, according to the above command, new problems appeared

zerollzeng commented 10 months ago

Which JP/DOS version you are using?

ou525 commented 10 months ago

image @zerollzeng

zerollzeng commented 9 months ago

I didn't test on JP 5.1. but JP 6.0 should work, could you please upgrade to it?

ou525 commented 9 months ago

By replacing the nvsci header file with version 6.0, I have successfully compiled and run

zerollzeng commented 9 months ago

Glad you solve it, I'm closing this.

HelloChang1 commented 7 months ago

By replacing the nvsci header file with version 6.0, I have successfully compiled and run

hai，where can I find the the nvsci header file with version 6.0? I only find in https://developer.nvidia.com/embedded/jetson-linux-r351 ,but it doesn't work.